Five-point errors require changes to 8-33% of reports to remove positive heat slopes in three coral samples from 2003-2019

We tested the minimum number of bounded report changes needed to remove a positive coral heat slope. Four Atlantic and Pacific models used 9,623 Reef Check surveys in 2003-2019, with country, year, month and depth controls. With errors limited to five percentage points per report, the later Atlantic minimum was 61 of 778 reports (7.84%). Early and late Pacific minima were 832 of 3,470 (23.98%) and 1,340 of 4,052 (33.07%). The early Atlantic slope could not reach zero. On repeated-year sites with site effects, the later Atlantic minimum was 12 of 513. A same-sample country model gave 13, so sample selection matters. Exact sorting bounds, binary optimization and attainable outcome assignments agree. These are fixed-record worst-case sensitivity results, not evidence that errors occurred. They have no sampling coverage guarantee and do not establish a causal heat effect. Missing visits, selection, shared errors and surface heat limit use. Code, panel data, raw hashes and reproduction instructions are included.

coral bleaching · heat stress · observation bias · bounded error · sparse errors · sensitivity analysis